Abstract: One embodiment of the present invention provides a system that identifies a failure mechanism for a component. The system first constructs a library of failure signatures from a set of components, wherein each failure signature in the library characterizes a known failure mechanism associated with a component in the set of components. Next, the system receives a telemetry signal which is measured from a component under surveillance. The system then determines if there is an anomaly in the telemetry signal. If so, the system performs pattern-matching operations between the telemetry signal and failure signatures in the library. Next, if a match is found, the system identifies the failure mechanism for the component under surveillance based on the matching failure signature.

Abstract: A system that reduces the size of a design data set. During this design data set reduction operation, the system computes a decision boundary which separates a first group of data patterns in a training data set from a second group of data patterns in the training data set. For each data pattern in the training data set, the system determines if removing the data pattern from the training data set substantially affects the resulting decision boundary. If so, the system marks the data pattern as a key pattern. The system then removes all data patterns that are not marked as key patterns to produce a reduced training data set which represents the decision boundary.

Abstract: A system that detects a change point in a time series of telemetry signals from a computer system. During operation, the system receives the time series of telemetry signals from the computer system. For each element in the time series, the system (1) inserts the element into a data structure which keeps track of the number of elements in the data structure that have a value greater than and that have a value less than the value of the inserted element; and (2) uses the information stored in the data structure to add a contribution by the inserted element to a trend statistic for the time series. The system then uses the trend statistic to select a hypothesis for the trend in the time series.

Abstract: In one embodiment, a method and apparatus for high-efficiency time-series archiving for computer server telemetry signals is disclosed. The method includes selecting one or more telemetry signals of a plurality of telemetry signals by a sequential probability ratio test (SPRT) algorithm, the SPRT algorithm identifying the one or more telemetry signals as not consistent with normal behavior of the plurality of telemetry signals, injecting synthetic samples around the selected one or more telemetry signals to create a continuous time series telemetry sample, and analyzing the continuous time series telemetry sample to identify leading indicators of faults in the target. Other embodiments are also disclosed.

Abstract: One embodiment of the present invention provides a system that determines a severity of degradation in a signal. During operation, the system receives signal values for the signal, wherein the signal values are received with a constant sampling interval. Next, for each received signal value, the system applies a Sequential Probability Ratio Test (SPRT) to the signal value. If the SPRT generates an alarm on the signal value, the system increments a cumulative counter which records a running total number of the SPRT alarms. Upon receiving each signal value, the system updates a cumulative function using a value in the cumulative counter. Next, the system determines the severity of degradation in the signal from the shape of the cumulative function.

Abstract: One embodiment of the present invention provides a system that optimizes a regression model which predicts a signal as a function of a set of available signals. During operation, the system receives training data for the set of available signals from a computer system during normal fault-free operation. The system also receives an objective function which can be used to evaluate how well a regression model predicts the signal. Next, the system initializes a pool of candidate regression models which includes at least two candidate regression models, wherein each candidate regression model in the pool includes a subset of the set of available signals.

Abstract: A system that provides fault tolerance in a parallel processing system. During operation, the system executes a parallel computing application in parallel across a subset of computing nodes within the parallel processing system. During this process, the system monitors telemetry signals within the parallel processing system. The system analyzes the monitored telemetry signals to determine if the probability that the parallel processing system will fail is increasing. If so, the system increases the frequency at which the parallel computing application is checkpointed, wherein a checkpoint includes the state of the parallel computing application at each computing node within the parallel processing system.

Abstract: One embodiment of the present invention provides a system that enhances throughput and fault-tolerance in a parallel-processing system. During operation, the system first receives a task. Next, the system partitions N computing nodes into M set-aside nodes and N-M primary computing nodes, wherein M?1. The system then processes the task in parallel across the N-M primary computing nodes. While doing so, the system proactively monitors the health of each of the N-M primary computing nodes. If the system detects a node in the N-M primary computing nodes to be at risk of failure, the system copies the portion of the task associated with the at-risk node to a subset of the M set-aside nodes. The system then processes the portion of the task in parallel across the subset of the M set-aside nodes while the N-M primary computing nodes continue executing.

Abstract: One embodiment of the present invention provides a system that tests the quality and/or the reliability of a component. During operation, the system applies test conditions to a plurality of specimens of the component. While applying the test conditions, the system measures the same variable from each of the plurality of specimens. Next, the system computes a running average of the measured variable across the plurality of specimens. The system then computes residuals between the measured variable for each specimen and the running average. The system next determines from the residuals whether the associated specimens are degraded.

Abstract: One embodiment of the present invention provides a system that facilitates high-sensitivity detection of an anomaly in a signal, wherein the signal is sampled to produce a set of possible quantized signal values. During operation, the system constructs a “reference distribution” for an “occurrence frequency” of a specific quantized signal value from the set of possible quantized signal values. The system then obtains a “deviant distribution” associated with the reference distribution, wherein the deviant distribution has an offset from the reference distribution to indicate an anomaly in the signal. Next, in response to a new occurrence of the specific quantized signal value, the system updates a mean and a variance of the reference distribution for the specific quantized signal value. The system also adjusts the deviant distribution for the specific quantized signal value based on the updated mean and the updated variance of the reference distribution for the specific quantized signal value.

Abstract: A system that reduces execution time of a parallel SVM application. During operation, the system partitions an input data set into chunks of data. Next, the system distributes the partitioned chunks of data across a plurality of available computing nodes and executes the parallel SVM application on the chunks of data in parallel across the plurality of available computing nodes. The system then determines if a first timeout period has been exceeded before all of the plurality of available computing nodes have finished processing their respective chunks of data. If so, the system (1) repartitions the input data set into different chunks of data; (2) redistributes the repartitioned chunks of data across some or all of the plurality of available computing nodes; and (3) executes the parallel SVM application on the repartitioned chunks of data in parallel across some or all of the available computing nodes.

Abstract: One embodiment of the present invention provides a system that reconstructs a high-resolution signal from a set of low-resolution quantized samples. During operation, the system receives a time series containing low-resolution quantized signal values which are sampled from the high-resolution signal. Next, the system performs a spectral analysis on the time series to obtain a frequency series for the low-resolution quantized signal values. The system next selects a subset of frequency terms from the frequency series which have the largest amplitudes. The system then reconstructs the high-resolution signal by performing an inverse spectral analysis on the subset of the frequency terms.

Abstract: A system that uses statistical techniques to selectively transmit data from a sensor. During operation, the system receives a sequence of quantized values from the sensor. The system then determines whether a distribution for the sequence of quantized values indicates that the sensor is observing a real event. If so, the system transmits sensor data for the real event to a receiver.

Abstract: A system that facilitates estimating power consumption in a computer system by inferring the power consumption from instrumentation signals. During operation, the system monitors instrumentation signals within the computer system, wherein the instrumentation signals do not include corresponding current and voltage signals that can be used to directly compute power consumption. The system then estimates the power consumption for the computer system by inferring the power consumption from the instrumentation signals and from an inferential power model generated during a training phase.

Abstract: One embodiment of the present invention provides a system for generating telemetric impulsional response fingerprints for an electronic system. The system operates by first determining a steady-state response of the electronic system under specified initial conditions. Next, the system introduces a sudden impulse step change to a parameter of the electronic system and then measures the dynamic response of the electronic system to the sudden impulse step change. The system then generates a multiparametric representation from the steady-state response and the dynamic response wherein the multiparametric representation simultaneously displays the steady-state response and the dynamic response.

Abstract: A system that identifies processes with a memory leak in a computer system. During operation, the system periodically samples memory usage for processes running on the computer system. The system then ranks the processes by memory usage and selects a specified number of processes with highest memory usage based on the ranking. For each selected process, the system computes a first-order difference of memory usage by taking a difference between the memory usage at a current sampling time and the memory usage at an immediately preceding sampling time. The system then generates a memory-leak index based on the first-order difference and a preceding memory-leak index computed at the immediately preceding sampling time.

Abstract: A system that facilitates reducing uncertainty in a quantized signal. During operation, the system measures a quantized output signal from a sensor. Next, the system obtains an initial value for an uncertainty interval for the quantized output signal. The system then margins the quantized output signal high by introducing a controlled increase in the mean of the quantized output signal to produce a high-margined quantized output signal. Next, the system measures the high-margined quantized output signal from the sensor. The system then uses information obtained from the high-margined quantized output signal to reduce the uncertainty interval for the quantized output signal.

Abstract: A system that detects the onset of degradation for interconnections in a component within a computer system. During operation, the system monitors inferential variables associated with the interconnections during operation of the computer system. Next, the system determines a present state of the component from the monitored inferential variables. The system then compares the present state of the component with an initial state of the component. If the comparison indicates that the interconnections in the component have reached or will reach a limited operating state (LOS), the system performs a remedial action.

Abstract: A system that detects the onset of hard disk drive failure. During operation, the system measures vibrations from the hard disk drive to produce one or more vibration signals. Next, the system generates a vibration signature for the hard disk drive from the measured vibration signals. The system then determines if the vibration signature indicates the onset of hard disk failure by comparing the vibration signature with a reference vibration signature for the hard disk drive. If so, the system generates a warning or takes a remedial action.

Abstract: One embodiment of the present invention provides a system that systematically monitors and records performance parameters for a computer system. During operation, the system periodically measures values for a set of performance parameters associated with the computer system while the computer system continues operating. The system then records the values on a data storage device, wherein the recording process keeps track of temporal relationships between events in different performance parameters. The system subsequently allows the recorded values for the set of performance parameters to be analyzed.